fs2.kafka.KafkaConsumer.scala Maven / Gradle / Ivy
The newest version!
/*
* Copyright 2018-2024 OVO Energy Limited
*
* SPDX-License-Identifier: Apache-2.0
*/
package fs2.kafka
import java.util
import scala.annotation.nowarn
import scala.collection.immutable.SortedSet
import scala.concurrent.duration.FiniteDuration
import scala.util.matching.Regex
import cats.{Foldable, Functor, Reducible}
import cats.data.{NonEmptySet, OptionT}
import cats.effect.*
import cats.effect.implicits.*
import cats.effect.std.*
import cats.syntax.all.*
import fs2.{Chunk, Stream}
import fs2.kafka.consumer.*
import fs2.kafka.consumer.KafkaConsumeChunk.CommitNow
import fs2.kafka.instances.*
import fs2.kafka.internal.*
import fs2.kafka.internal.converters.collection.*
import fs2.kafka.internal.syntax.*
import fs2.kafka.internal.KafkaConsumerActor.*
import fs2.kafka.internal.LogEntry.{RevokedPreviousFetch, StoredFetch}
import org.apache.kafka.clients.consumer.{OffsetAndMetadata, OffsetAndTimestamp}
import org.apache.kafka.common.{Metric, MetricName, PartitionInfo, TopicPartition}
/**
* [[KafkaConsumer]] represents a consumer of Kafka records, with the ability to `subscribe` to
* topics, start a single top-level stream, and optionally control it via the provided [[fiber]]
* instance.
*
* The following top-level streams are provided.
* - [[stream]] provides a single stream of records, where the order of records is guaranteed per
* topic-partition.
* - [[partitionedStream]] provides a stream with elements as streams that continually request
* records for a single partition. Order is guaranteed per topic-partition, but all assigned
* partitions will have to be processed in parallel.
* - [[partitionsMapStream]] provides a stream where each element contains a current assignment.
* The current assignment is the `Map`, where keys is a `TopicPartition`, and values are
* streams with records for a particular `TopicPartition`.
For the streams, records are
* wrapped in [[CommittableConsumerRecord]]s which provide [[CommittableOffset]]s with the
* ability to commit record offsets to Kafka. For performance reasons, offsets are usually
* committed in batches using [[CommittableOffsetBatch]]. Provided `Pipe`s, like
* [[commitBatchWithin]] are available for batch committing offsets. If you are not committing
* offsets to Kafka, you can simply discard the [[CommittableOffset]], and only make use of the
* record.
*
* While it's technically possible to start more than one stream from a single [[KafkaConsumer]],
* it is generally not recommended as there is no guarantee which stream will receive which
* records, and there might be an overlap, in terms of duplicate records, between the two streams.
* If a first stream completes, possibly with error, there's no guarantee the stream has processed
* all of the records it received, and a second stream from the same [[KafkaConsumer]] might not be
* able to pick up where the first one left off. Therefore, only create a single top-level stream
* per [[KafkaConsumer]], and if you want to start a new stream if the first one finishes, let the
* [[KafkaConsumer]] shutdown and create a new one.
*/
sealed abstract class KafkaConsumer[F[_], K, V]
extends KafkaConsume[F, K, V]
with KafkaConsumeChunk[F, K, V]
with KafkaAssignment[F]
with KafkaOffsetsV2[F]
with KafkaSubscription[F]
with KafkaTopicsV2[F]
with KafkaCommit[F]
with KafkaMetrics[F]
with KafkaConsumerLifecycle[F]
object KafkaConsumer {
/**
* Processes requests from the queue, if there are pending requests, otherwise waits for the next
* poll.
*
* In particular, any newly queued requests may wait for up to pollInterval, and for the next
* poll to complete.
*
* The resulting effect runs forever, until canceled.
*/
private def runConsumerActor[F[_], K, V](
requests: QueueSource[F, Request[F, K, V]],
polls: QueueSource[F, Request.Poll[F]],
actor: KafkaConsumerActor[F, K, V]
)(implicit
F: Async[F]
): F[Unit] =
OptionT(requests.tryTake).getOrElseF(polls.take.widen).flatMap(actor.handle).foreverM[Unit]
/**
* Schedules polls every pollInterval to be handled by runConsumerActor.
*
* The polls queue is assumed bounded to provide backpressure.
*
* The resulting effect runs forever, until canceled.
*/
private def runPollScheduler[F[_], K, V](
polls: QueueSink[F, Request.Poll[F]],
pollInterval: FiniteDuration
)(implicit
F: Temporal[F]
): F[Unit] =
polls.offer(Request.poll).andWait(pollInterval).foreverM[Unit]
private def startBackgroundConsumer[F[_], K, V](
requests: QueueSource[F, Request[F, K, V]],
polls: Queue[F, Request.Poll[F]],
actor: KafkaConsumerActor[F, K, V],
pollInterval: FiniteDuration
)(implicit
F: Async[F]
): Resource[F, Fiber[F, Throwable, Unit]] =
Resource.make {
F.race(
runConsumerActor(requests, polls, actor),
runPollScheduler(polls, pollInterval)
)
.void
.start
}(_.cancel.start.void)
private def createKafkaConsumer[F[_], K, V](
requests: QueueSink[F, Request[F, K, V]],
settings: ConsumerSettings[F, K, V],
actor: KafkaConsumerActor[F, K, V],
fiber: Fiber[F, Throwable, Unit],
streamIdRef: Ref[F, StreamId],
id: Int,
withConsumer: WithConsumer[F],
stopConsumingDeferred: Deferred[F, Unit]
)(implicit F: Async[F], logging: Logging[F]): KafkaConsumer[F, K, V] =
new KafkaConsumer[F, K, V] {
override def partitionsMapStream
: Stream[F, Map[TopicPartition, Stream[F, CommittableConsumerRecord[F, K, V]]]] = {
val chunkQueue: F[Queue[F, Option[Chunk[CommittableConsumerRecord[F, K, V]]]]] =
Queue.bounded(settings.maxPrefetchBatches - 1)
type PartitionResult =
(Chunk[CommittableConsumerRecord[F, K, V]], FetchCompletedReason)
type PartitionsMap = Map[TopicPartition, Stream[F, CommittableConsumerRecord[F, K, V]]]
type PartitionsMapQueue = Queue[F, Option[PartitionsMap]]
def partitionStream(
streamId: StreamId,
partition: TopicPartition,
assignmentRevoked: F[Unit]
): Stream[F, CommittableConsumerRecord[F, K, V]] = Stream.force {
for {
chunks <- chunkQueue
dequeueDone <- Deferred[F, Unit]
shutdown = F.race(
F.race(
awaitTermination.attempt,
dequeueDone.get
),
F.race(
stopConsumingDeferred.get,
assignmentRevoked
)
)
.void
stopReqs <- Deferred[F, Unit]
} yield Stream
.eval {
val fetchPartition: F[Unit] = F
.deferred[PartitionResult]
.flatMap { deferred =>
val callback: PartitionResult => F[Unit] =
deferred.complete(_).void
val fetch: F[PartitionResult] = withPermit {
val assigned =
withConsumer.blocking {
_.assignment.contains(partition)
}
def storeFetch: F[Unit] =
actor
.ref
.modify { state =>
val (newState, oldFetches) =
state.withFetch(partition, streamId, callback)
newState ->
(logging.log(StoredFetch(partition, callback, newState)) >>
oldFetches.traverse_ { fetch =>
fetch.completeRevoked(Chunk.empty) >>
logging.log(RevokedPreviousFetch(partition, streamId))
})
}
.flatten
def completeRevoked: F[Unit] =
callback((Chunk.empty, FetchCompletedReason.TopicPartitionRevoked))
assigned.ifM(storeFetch, completeRevoked)
} >> deferred.get
fetch.flatMap { case (chunk, reason) =>
val enqueueChunk = chunks.offer(Some(chunk)).unlessA(chunk.isEmpty)
val completeRevoked =
stopReqs.complete(()).void.whenA(reason.topicPartitionRevoked)
enqueueChunk >> completeRevoked
}
}
Stream
.repeatEval {
stopReqs
.tryGet
.flatMap {
case None =>
fetchPartition
case Some(()) =>
// Prevent issuing additional requests after partition is
// revoked or shutdown happens, in case the stream isn't
// interrupted fast enough
F.unit
}
}
.interruptWhen(F.race(shutdown, stopReqs.get).void.attempt)
.compile
.drain
.guarantee(F.race(dequeueDone.get, chunks.offer(None)).void)
.start
.as {
Stream
.fromQueueNoneTerminated(chunks)
.flatMap(Stream.chunk)
.covary[F]
.onFinalize(dequeueDone.complete(()).void)
}
}
.flatten
}
def enqueueAssignment(
streamId: StreamId,
assigned: Map[TopicPartition, Deferred[F, Unit]],
partitionsMapQueue: PartitionsMapQueue
): F[Unit] =
stopConsumingDeferred
.tryGet
.flatMap {
case None =>
val assignment: PartitionsMap = assigned.map { case (partition, finisher) =>
partition -> partitionStream(streamId, partition, finisher.get)
}
partitionsMapQueue.offer(Some(assignment))
case Some(()) =>
F.unit
}
def onRebalance(
streamId: StreamId,
assignmentRef: Ref[F, Map[TopicPartition, Deferred[F, Unit]]],
partitionsMapQueue: PartitionsMapQueue
): OnRebalance[F] =
OnRebalance(
onRevoked = revoked => {
for {
finishers <- assignmentRef.modify(_.partition(entry => !revoked.contains(entry._1)))
_ <- finishers.toVector.traverse { case (_, finisher) => finisher.complete(()) }
} yield ()
},
onAssigned = assignedPartitions => {
for {
assignment <- assignedPartitions
.toVector
.traverse { partition =>
Deferred[F, Unit].map(partition -> _)
}
.map(_.toMap)
_ <- assignmentRef.update(_ ++ assignment)
_ <- enqueueAssignment(
streamId = streamId,
assigned = assignment,
partitionsMapQueue = partitionsMapQueue
)
} yield ()
}
)
def requestAssignment(
streamId: StreamId,
assignmentRef: Ref[F, Map[TopicPartition, Deferred[F, Unit]]],
partitionsMapQueue: PartitionsMapQueue
): F[Map[TopicPartition, Deferred[F, Unit]]] = {
val assignment = this.assignment(
Some(
onRebalance(
streamId,
assignmentRef,
partitionsMapQueue
)
)
)
F.race(awaitTermination.attempt, assignment)
.flatMap {
case Left(_) =>
F.pure(Map.empty)
case Right(assigned) =>
assigned
.toVector
.traverse { partition =>
Deferred[F, Unit].map(partition -> _)
}
.map(_.toMap)
}
}
def initialEnqueue(
streamId: StreamId,
assignmentRef: Ref[F, Map[TopicPartition, Deferred[F, Unit]]],
partitionsMapQueue: PartitionsMapQueue
): F[Unit] =
for {
assigned <- requestAssignment(
streamId,
assignmentRef,
partitionsMapQueue
)
_ <- enqueueAssignment(streamId, assigned, partitionsMapQueue)
} yield ()
Stream
.eval(stopConsumingDeferred.tryGet)
.flatMap {
case None =>
for {
partitionsMapQueue <- Stream.eval(Queue.unbounded[F, Option[PartitionsMap]])
streamId <- Stream.eval(streamIdRef.modify(n => (n + 1, n)))
assignmentRef <- Stream
.eval(Ref[F].of(Map.empty[TopicPartition, Deferred[F, Unit]]))
_ <- Stream.eval(
initialEnqueue(
streamId,
assignmentRef,
partitionsMapQueue
)
)
out <- Stream
.fromQueueNoneTerminated(partitionsMapQueue)
.interruptWhen(awaitTermination.attempt)
.concurrently(
Stream.eval(stopConsumingDeferred.get >> partitionsMapQueue.offer(None))
)
} yield out
case Some(()) =>
Stream.empty.covaryAll[F, PartitionsMap]
}
}
override def partitionedStream: Stream[F, Stream[F, CommittableConsumerRecord[F, K, V]]] =
partitionsMapStream.flatMap(partitionsMap => Stream.iterable(partitionsMap.values))
override def stream: Stream[F, CommittableConsumerRecord[F, K, V]] =
partitionedStream.parJoinUnbounded
override def commitAsync(offsets: Map[TopicPartition, OffsetAndMetadata]): F[Unit] =
request { callback =>
Request.ManualCommitAsync(
callback = callback,
offsets = offsets
)
}
override def commitSync(offsets: Map[TopicPartition, OffsetAndMetadata]): F[Unit] =
request { callback =>
Request.ManualCommitSync(
callback = callback,
offsets = offsets
)
}
private[this] def request[A](
request: (Either[Throwable, A] => F[Unit]) => Request[F, K, V]
): F[A] =
Deferred[F, Either[Throwable, A]]
.flatMap { deferred =>
requests.offer(request(deferred.complete(_).void)) >>
F.race(awaitTermination.as(ConsumerShutdownException()), deferred.get.rethrow)
}
.rethrow
override def assignment: F[SortedSet[TopicPartition]] =
assignment(Option.empty)
private def assignment(
onRebalance: Option[OnRebalance[F]]
): F[SortedSet[TopicPartition]] =
withPermit {
onRebalance
.fold(actor.ref.updateAndGet(_.asStreaming)) { on =>
actor
.ref
.updateAndGet(_.withOnRebalance(on).asStreaming)
.flatTap { newState =>
logging.log(LogEntry.StoredOnRebalance(on, newState))
}
}
.ensure(NotSubscribedException())(_.subscribed) >>
withConsumer.blocking(_.assignment.toSortedSet)
}
override def assignmentStream: Stream[F, SortedSet[TopicPartition]] = {
// NOTE: `initialAssignmentDone` is needed here to guard against the
// race condition when a rebalance triggers after the listeners are
// registered but before `assignmentRef` can be updated with initial
// assignments.
def onRebalanceWith(
updateQueue: Queue[F, SortedSet[TopicPartition]],
assignmentRef: Ref[F, SortedSet[TopicPartition]],
initialAssignmentDone: F[Unit]
): OnRebalance[F] =
OnRebalance(
onAssigned = assigned =>
initialAssignmentDone >>
assignmentRef.updateAndGet(_ ++ assigned).flatMap(updateQueue.offer),
onRevoked = revoked =>
initialAssignmentDone >>
assignmentRef.updateAndGet(_ -- revoked).flatMap(updateQueue.offer)
)
Stream
.eval {
(
Queue.unbounded[F, SortedSet[TopicPartition]],
Ref[F].of(SortedSet.empty[TopicPartition]),
Deferred[F, Unit]
).tupled
.flatMap[Stream[F, SortedSet[TopicPartition]]] {
case (updateQueue, assignmentRef, initialAssignmentDeferred) =>
val onRebalance =
onRebalanceWith(
updateQueue = updateQueue,
assignmentRef = assignmentRef,
initialAssignmentDone = initialAssignmentDeferred.get
)
assignment(Some(onRebalance))
.flatMap { initialAssignment =>
assignmentRef.set(initialAssignment) >>
updateQueue.offer(initialAssignment) >>
initialAssignmentDeferred.complete(())
}
.as(Stream.fromQueueUnterminated(updateQueue).changes)
}
}
.flatten
}
override def seek(partition: TopicPartition, offset: Long): F[Unit] =
withConsumer.blocking(_.seek(partition, offset))
override def seekToBeginning[G[_]](partitions: G[TopicPartition])(implicit
G: Foldable[G]
): F[Unit] =
withConsumer.blocking(_.seekToBeginning(partitions.asJava))
override def seekToEnd[G[_]](
partitions: G[TopicPartition]
)(implicit G: Foldable[G]): F[Unit] =
withConsumer.blocking(_.seekToEnd(partitions.asJava))
override def partitionsFor(
topic: String
): F[List[PartitionInfo]] =
withConsumer.blocking(_.partitionsFor(topic).asScala.toList)
override def partitionsFor(
topic: String,
timeout: FiniteDuration
): F[List[PartitionInfo]] =
withConsumer.blocking(_.partitionsFor(topic, timeout.toJava).asScala.toList)
override def position(partition: TopicPartition): F[Long] =
withConsumer.blocking(_.position(partition))
override def position(partition: TopicPartition, timeout: FiniteDuration): F[Long] =
withConsumer.blocking(_.position(partition, timeout.toJava))
override def committed(
partitions: Set[TopicPartition]
): F[Map[TopicPartition, OffsetAndMetadata]] =
withConsumer.blocking {
_.committed(partitions.asJava)
.asInstanceOf[util.Map[TopicPartition, OffsetAndMetadata]]
.toMap
}
override def committed(
partitions: Set[TopicPartition],
timeout: FiniteDuration
): F[Map[TopicPartition, OffsetAndMetadata]] =
withConsumer.blocking {
_.committed(partitions.asJava, timeout.toJava)
.asInstanceOf[util.Map[TopicPartition, OffsetAndMetadata]]
.toMap
}
override def subscribe[G[_]](topics: G[String])(implicit G: Reducible[G]): F[Unit] =
withPermit {
withConsumer.blocking {
_.subscribe(
topics.toList.asJava,
actor.consumerRebalanceListener
)
} >> actor
.ref
.updateAndGet(_.asSubscribed)
.log(LogEntry.SubscribedTopics(topics.toNonEmptyList, _))
}
private def withPermit[A](fa: F[A]): F[A] =
F.deferred[Either[Throwable, A]]
.flatMap { deferred =>
requests.offer(
Request.WithPermit(fa, deferred.complete(_: Either[Throwable, A]).void)
) >> deferred.get.rethrow
}
override def subscribe(regex: Regex): F[Unit] =
withPermit {
withConsumer.blocking {
_.subscribe(
regex.pattern,
actor.consumerRebalanceListener
)
} >>
actor.ref.updateAndGet(_.asSubscribed).log(LogEntry.SubscribedPattern(regex.pattern, _))
}
override def unsubscribe: F[Unit] =
withPermit {
withConsumer.blocking(_.unsubscribe()) >> actor
.ref
.updateAndGet(_.asUnsubscribed)
.log(LogEntry.Unsubscribed(_))
}
override def stopConsuming: F[Unit] =
stopConsumingDeferred.complete(()).attempt.void
override def assign(partitions: NonEmptySet[TopicPartition]): F[Unit] =
withPermit {
withConsumer.blocking {
_.assign(
partitions.toList.asJava
)
} >> actor
.ref
.updateAndGet(_.asSubscribed)
.log(LogEntry.ManuallyAssignedPartitions(partitions, _))
}
override def assign(topic: String): F[Unit] =
for {
partitions <- partitionsFor(topic).map { partitionInfo =>
NonEmptySet.fromSet {
SortedSet(partitionInfo.map(_.partition)*)
}
}
_ <- partitions.fold(F.unit)(assign(topic, _))
} yield ()
override def beginningOffsets(
partitions: Set[TopicPartition]
): F[Map[TopicPartition, Long]] =
withConsumer.blocking {
_.beginningOffsets(partitions.asJava).asInstanceOf[util.Map[TopicPartition, Long]].toMap
}
override def beginningOffsets(
partitions: Set[TopicPartition],
timeout: FiniteDuration
): F[Map[TopicPartition, Long]] =
withConsumer.blocking {
_.beginningOffsets(partitions.asJava, timeout.toJava)
.asInstanceOf[util.Map[TopicPartition, Long]]
.toMap
}
override def endOffsets(
partitions: Set[TopicPartition]
): F[Map[TopicPartition, Long]] =
withConsumer.blocking {
_.endOffsets(partitions.asJava).asInstanceOf[util.Map[TopicPartition, Long]].toMap
}
override def endOffsets(
partitions: Set[TopicPartition],
timeout: FiniteDuration
): F[Map[TopicPartition, Long]] =
withConsumer.blocking {
_.endOffsets(partitions.asJava, timeout.toJava)
.asInstanceOf[util.Map[TopicPartition, Long]]
.toMap
}
override def offsetsForTimes(
timestampsToSearch: Map[TopicPartition, Long]
): F[Map[TopicPartition, Option[OffsetAndTimestamp]]] =
withConsumer.blocking {
_.offsetsForTimes(
timestampsToSearch.asJava.asInstanceOf[util.Map[TopicPartition, java.lang.Long]]
)
// Convert empty/missing partition null values to None for more idiomatic scala
.toMapOptionValues
}
override def offsetsForTimes(
timestampsToSearch: Map[TopicPartition, Long],
timeout: FiniteDuration
): F[Map[TopicPartition, Option[OffsetAndTimestamp]]] =
withConsumer.blocking {
_.offsetsForTimes(
timestampsToSearch.asJava.asInstanceOf[util.Map[TopicPartition, java.lang.Long]],
timeout.toJava
)
// Convert empty/missing partition null values to None for more idiomatic scala
.toMapOptionValues
}
override def listTopics: F[Map[String, List[PartitionInfo]]] =
withConsumer.blocking {
_.listTopics().toMap.map { case (k, v) => (k, v.toList) }
}
override def listTopics(timeout: FiniteDuration): F[Map[String, List[PartitionInfo]]] =
withConsumer.blocking {
_.listTopics(timeout.toJava).toMap.map { case (k, v) => (k, v.toList) }
}
override def metrics: F[Map[MetricName, Metric]] =
withConsumer.blocking(_.metrics().asScala.toMap)
override def toString: String =
"KafkaConsumer$" + id
override def terminate: F[Unit] = fiber.cancel.start.void
override def awaitTermination: F[Unit] = fiber.joinWithUnit
}
/**
* Creates a new [[KafkaConsumer]] in the `Resource` context, using the specified
* [[ConsumerSettings]]. Note that there is another version where `F[_]` is specified explicitly
* and the key and value type can be inferred, which allows you to use the following syntax.
*
* {{{
* KafkaConsumer.resource[F].using(settings)
* }}}
*/
def resource[F[_], K, V](
settings: ConsumerSettings[F, K, V]
)(implicit
F: Async[F],
mk: MkConsumer[F]
): Resource[F, KafkaConsumer[F, K, V]] =
for {
keyDeserializer <- settings.keyDeserializer
valueDeserializer <- settings.valueDeserializer
id <- Resource.eval(F.delay(new Object().hashCode))
jitter <- Resource.eval(Jitter.default[F])
logging <- Resource.eval(Logging.default[F](id))
requests <- Resource.eval(Queue.unbounded[F, Request[F, K, V]])
polls <- Resource.eval(Queue.bounded[F, Request.Poll[F]](1))
ref <- Resource.eval(Ref.of[F, State[F, K, V]](State.empty))
streamId <- Resource.eval(Ref.of[F, StreamId](0))
dispatcher <- Dispatcher.sequential[F]
stopConsumingDeferred <- Resource.eval(Deferred[F, Unit])
withConsumer <- WithConsumer(mk, settings)
actor = {
implicit val jitter0: Jitter[F] = jitter
implicit val logging0: Logging[F] = logging
implicit val dispatcher0: Dispatcher[F] = dispatcher
new KafkaConsumerActor(
settings = settings,
keyDeserializer = keyDeserializer,
valueDeserializer = valueDeserializer,
ref = ref,
requests = requests,
withConsumer = withConsumer
)
}
fiber <- startBackgroundConsumer(requests, polls, actor, settings.pollInterval)
} yield createKafkaConsumer(
requests,
settings,
actor,
fiber,
streamId,
id,
withConsumer,
stopConsumingDeferred
)(F, logging)
/**
* Creates a new [[KafkaConsumer]] in the `Stream` context, using the specified
* [[ConsumerSettings]]. Note that there is another version where `F[_]` is specified explicitly
* and the key and value type can be inferred, which allows you to use the following syntax.
*
* {{{
* KafkaConsumer.stream[F].using(settings)
* }}}
*/
def stream[F[_], K, V](
settings: ConsumerSettings[F, K, V]
)(implicit F: Async[F], mk: MkConsumer[F]): Stream[F, KafkaConsumer[F, K, V]] =
Stream.resource(resource(settings)(F, mk))
def apply[F[_]]: ConsumerPartiallyApplied[F] =
new ConsumerPartiallyApplied()
final private[kafka] class ConsumerPartiallyApplied[F[_]](val dummy: Boolean = true)
extends AnyVal {
/**
* Alternative version of `resource` where the `F[_]` is specified explicitly, and where the
* key and value type can be inferred from the [[ConsumerSettings]]. This allows you to use the
* following syntax.
*
* {{{
* KafkaConsumer[F].resource(settings)
* }}}
*/
def resource[K, V](settings: ConsumerSettings[F, K, V])(implicit
F: Async[F],
mk: MkConsumer[F]
): Resource[F, KafkaConsumer[F, K, V]] =
KafkaConsumer.resource(settings)(F, mk)
/**
* Alternative version of `stream` where the `F[_]` is specified explicitly, and where the key
* and value type can be inferred from the [[ConsumerSettings]]. This allows you to use the
* following syntax.
*
* {{{
* KafkaConsumer[F].stream(settings)
* }}}
*/
def stream[K, V](settings: ConsumerSettings[F, K, V])(implicit
F: Async[F],
mk: MkConsumer[F]
): Stream[F, KafkaConsumer[F, K, V]] =
KafkaConsumer.stream(settings)(F, mk)
override def toString: String =
"ConsumerPartiallyApplied$" + System.identityHashCode(this)
}
/*
* Extension methods for operating on a `KafkaConsumer` in a `Stream` context without needing
* to explicitly use operations such as `flatMap` and `evalTap`
*/
implicit final class StreamOps[F[_]: Functor, K, V](self: Stream[F, KafkaConsumer[F, K, V]]) {
/**
* Subscribes a consumer to the specified topics within the [[Stream]] context. See
* [[KafkaSubscription#subscribe]].
*/
def subscribe[G[_]: Reducible](topics: G[String]): Stream[F, KafkaConsumer[F, K, V]] =
self.evalTap(_.subscribe(topics))
def subscribe(regex: Regex): Stream[F, KafkaConsumer[F, K, V]] =
self.evalTap(_.subscribe(regex))
/**
* Subscribes a consumer to the specified topics within the [[Stream]] context. See
* [[KafkaSubscription#subscribe]].
*/
def subscribeTo(
firstTopic: String,
remainingTopics: String*
): Stream[F, KafkaConsumer[F, K, V]] =
self.evalTap(_.subscribeTo(firstTopic, remainingTopics*))
/**
* A [[Stream]] of records from the allocated [[KafkaConsumer]]. Alias for [[stream]]. See
* [[KafkaConsume#stream]]
*/
def records: Stream[F, CommittableConsumerRecord[F, K, V]] = stream
/**
* A [[Stream]] of records from the allocated [[KafkaConsumer]]. See [[KafkaConsume#stream]]
*/
def stream: Stream[F, CommittableConsumerRecord[F, K, V]] = self.flatMap(_.records)
/**
* Alias for [[partitionedStream]]. See [[KafkaConsume#partitionedStream]]
*/
def partitionedRecords: Stream[F, Stream[F, CommittableConsumerRecord[F, K, V]]] =
partitionedStream
/**
* See [[KafkaConsume#partitionedStream]]
*/
def partitionedStream: Stream[F, Stream[F, CommittableConsumerRecord[F, K, V]]] =
self.flatMap(_.partitionedRecords)
/**
* Consume from all assigned partitions concurrently, processing the messages in `Chunk`s. See
* [[KafkaConsumeChunk#consumeChunk]]
*/
def consumeChunk(processor: Chunk[ConsumerRecord[K, V]] => F[CommitNow])(implicit
F: Concurrent[F]
): F[Nothing] = self.evalMap(_.consumeChunk(processor)).compile.onlyOrError
}
/*
* Prevents the default `MkConsumer` instance from being implicitly available
* to code defined in this object, ensuring factory methods require an instance
* to be provided at the call site.
*/
@nowarn("msg=never used")
implicit private def mkAmbig1[F[_]]: MkConsumer[F] =
throw new AssertionError("should not be used")
@nowarn("msg=never used")
implicit private def mkAmbig2[F[_]]: MkConsumer[F] =
throw new AssertionError("should not be used")
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy